When Tony Blair looked back on his time in power, he had a simple assessment of his decision to introduce the Freedom of Information Act: "You idiot." While the technology secretary, Peter Kyle, is a fan of the former prime minister, he may be inclined to agree with that verdict after the act was used to reveal that he had been asking ChatGPT which podcasts he should appear on. The disclosure has already caused frustration among ministers, given its possible repercussions. Blair's gripe was that the act risked stopping the frank discussions needed among ministers and officials. Ever since, it has become notoriously difficult to have a freedom of information (FoI) request granted, as officials exploit various legal exemptions to refuse them. The successful use of the legislation to probe into Kyle's AI chatbot use has led some to conclude that a new precedent has been set, one that will have officials across Whitehall sweating over their recent chatbot interactions.

Predictive models are often introduced to decision-making tasks under the rationale that they improve performance over an existing decision-making policy. However, it is challenging to compare predictive performance against an existing decision-making policy that is generally under-specified and dependent on unobservable factors. These sources of uncertainty are often addressed in practice by making strong assumptions about the data-generating mechanism. In this work, we propose a method to compare the predictive performance of decision policies under a variety of modern identification approaches from the causal inference and off-policy evaluation literatures (e.g., instrumental variable, marginal sensitivity model, proximal variable). Key to our method is the insight that there are regions of uncertainty that we can safely ignore in the policy comparison. We develop a practical approach for finite-sample estimation of regret intervals under no assumptions on the parametric form of the status quo policy. We verify our framework theoretically and via synthetic data experiments. We conclude with a real-world application using our framework to support a pre-deployment evaluation of a proposed modification to a healthcare enrollment policy.

In this work, we proposed a novel inferential procedure assisted by machine learning based adjustment for randomized control trials. The method was developed under the Rosenbaum's framework of exact tests in randomized experiments with covariate adjustments. Through extensive simulation experiments, we showed the proposed method can robustly control the type I error and can boost the inference efficiency for a randomized controlled trial (RCT). This advantage was further demonstrated in a real world example. The simplicity and robustness of the proposed method makes it a competitive candidate as a routine inference procedure for RCTs, especially when the number of baseline covariates is large, and when nonlinear association or interaction among covariates is expected. Its application may remarkably reduce the required sample size and cost of RCTs, such as phase III clinical trials.

Unobserved confounding is common in many applications, making causal inference from observational data challenging. As a remedy, causal sensitivity analysis is an important tool to draw causal conclusions under unobserved confounding with mathematical guarantees. In this paper, we propose NeuralCSA, a neural framework for generalized causal sensitivity analysis. Unlike previous work, our framework is compatible with (i) a large class of sensitivity models, including the marginal sensitivity model, f-sensitivity models, and Rosenbaum's sensitivity model; (ii) different treatment types (i.e., binary and continuous); and (iii) different causal queries, including (conditional) average treatment effects and simultaneous effects on multiple outcomes. The generality of \frameworkname is achieved by learning a latent distribution shift that corresponds to a treatment intervention using two conditional normalizing flows. We provide theoretical guarantees that NeuralCSA is able to infer valid bounds on the causal query of interest and also demonstrate this empirically using both simulated and real-world data.

If you're like me, the growing drip, drip, drip of the content faucet is pushing you to the edge: posts, pings, notifications, alerts. Tech journalist Charles Arthur makes a compelling argument that future content is at a tipping point. Arthur is the author of the substack blog "Social Warming," about social networks' effects on society. "The approaching tsunami of addictive AI-created content will overwhelm us" warns Arthur. The tsunami he points to is the creation of what academics call synthetic media, media that is created entirely by artificial intelligence.

We examine the role of textual data as study units when conducting causal inference by drawing parallels between human subjects and organized texts. We elaborate on key causal concepts and principles, and expose some ambiguity and sometimes fallacies. To facilitate better framing a causal query, we discuss two strategies: (i) shifting from immutable traits to perceptions of them, and (ii) shifting from some abstract concept/property to its constituent parts, i.e., adopting a constructivist perspective of an abstract concept. We hope this article would raise the awareness of the importance of articulating and clarifying fundamental concepts before delving into developing methodologies when drawing causal inference using textual data.

Defence lawyers in the Wisconsin murder trial of Kyle Rittenhouse said on Wednesday they would ask for a mistrial because of a dispute with prosecutors over video evidence, as the jury watched footage of his shootings at protests last year. Rittenhouse, 18, is charged with killing Joseph Rosenbaum, 36, and Anthony Huber, 26, and attempted homicide in the wounding of Gaige Grosskreutz, 28, during a chaotic night in Kenosha, Wisconsin, on August 25, 2020. The protests that night – marred by arson, rioting and looting – followed the police shooting of a Black man, Jacob Blake, who was left paralyzed from the waist down. Rittenhouse has pleaded not guilty. At issue in the trial is a drone video that shows Rosenbaum chasing Rittenhouse in the parking lot of a used-car dealership and the teenager turning and opening fire with his semi-automatic rifle as Rosenbaum gets close to him.

The jury at Kyle Rittenhouse's murder trial Tuesday watched drone video that showed Rittenhouse wheeling around and shooting Joseph Rosenbaum at close range during a night of turbulent protests on the streets of Kenosha. The video, zoomed in and slowed down by a forensic imaging specialist, was played as the prosecution's case appeared to be winding down after a week of testimony in which some of its own witnesses often bolstered Rittenhouse's claim of self-defense. The footage showed Rosenbaum following Rittenhouse before Rittenhouse suddenly spins around and fires his rifle at him. Rosenbaum falls, and Rittenhouse runs around a car. Dr. Doug Kelley, a forensic pathologist with the Milwaukee County medical examiner's office, said Rosenbaum was shot by someone who was within 4 feet of him.

Brainstorm guest contributor Paul Fraumeni speaks with four York U researchers who are applying artificial intelligence to their research ventures in ways that, ultimately, could lead to profound and positive impacts on health care in this country. Meet four York University researchers: Lauren Sergio and Doug Crawford have academic backgrounds in physiology; Shayna Rosenbaum has a PhD in psychology; Joel Zylberberg has a doctorate in physics. They share two things in common: They focus on neuroscience – the study of the brain and its functions – and they leverage advanced computing technology using artificial intelligence (AI) in their research ventures, the application of which could have a profound and positive impact on health care. In a nondescript room in the Sherman Health Sciences Research Centre, Lauren Sergio sits down and places her right arm in a sleeve on an armrest. It's an odd-looking contraption; the lower part looks like a sling attached to a video game joystick.

Rosenbaum: This is the video of a machine-learning simulation learning to walk and facing obstacles, and it's there only because I like it. Also, it's a kind of metaphor for me trying to build the CI/CD pipeline. I'm going to be talking about CI/CD for machine learning, which is also being called MLOps. The words are hard, we don't have to really define these things, but we do have to define some other things and we're going to talk about definitions a lot actually. I'm going to start by introducing myself. I'm on the left, this picture is from DevOpsDays Chicago, our mascot is a DevOps Yak. You can come check out the conference. I work for Microsoft on the Azure DevOps team. I come from a developer background, and then, I did a lot of things with DevOps CI/CD and such. I'm not a data scientist, I did some classes on machine learning just so I can get context on this, but I'm coming to this primarily from a developer perspective. I also run another conference, this is a shameless plug, it's DeliveryConf, it's the first year it's happening, it's going to be in Seattle, Washington, on January 21 and 22. You should register for it right now because it's going to be awesome. The first thing I want to do is I want to set an agenda.